Education and Voting Conservative: Evidencefromamajorschoolingreformin Great Britain

EDUCATION AND VOTING CONSERVATIVE: EVIDENCEFROMAMAJORSCHOOLINGREFORMIN GREAT BRITAIN

JOHN MARSHALL∗

FEBRUARY 2015

High school education is central to adolescent socialization, but also has important downstream consequences for adult life. However, scholars examining schooling’s political effects have struggled to reconcile education’s correlation with both more liberal social attitudes and greater income. To disentangle this relationship, I exploit a major school leaving age reform in Great Britain that caused almost half the population to remain at high school for at least an additional year. Using a fuzzy regression discontinuity design, I ﬁnd that each additional year of late high school increases the probability of voting Conservative in later life by 12 percentage points. A similar relationship holds away from the discontinuity, suggesting that education is a key determinant of voting behavior and that the reform could have signiﬁcantly altered electoral outcomes. I provide evidence suggesting that, by increasing an individual’s income, education increases support for right-wing economic policies, and ultimately the Conservative party.

∗PhD candidate, Department of Government, Harvard University. [email protected]. I thank Jim Alt, Charlotte Cavaille, Andy Hall, Torben Iversen, Horacio Larreguy, Brandon Stewart, and Tess Wise for illuminating discussions and useful comments.

1 1 Introduction

High school is a deﬁning feature of an individual’s adolescence and has been linked to radically different life trajectories. Such education may permanently instill social and political attitudes, determine labor market prospects, or affect the composition of an individual’s social network. Consequently, it has the potential to substantially alter a voter’s political preferences and voting behavior in later life, and in turn impact electoral and policy outcomes. However, despite considerable interest in education’s effect on political participation (see Sond- heimer and Green 2010), strikingly little is known about education’s effect on the party an individual votes for. Existing survey evidence, which has struggled to square the widely-documented correlations between income (which education increases) and support for conservative economic policies (e.g. Clarke et al. 2004; Gelman et al. 2010) and between education and socially liberal attitudes (e.g. Converse 1972; Nie, Junn and Stehlik-Barry 1996; Gerber et al. 2010), has failed to disentangle either the direction of the relationship or its mechanisms. In part, this reﬂects major empirical challenges stemming from the fact that better educated individuals differ substantially in other important respects and that education is itself a cause of many variables that researchers often control for. Furthermore, because the direct link to vote choice has received limited attention, it is also possible that education affects attitudes without impacting vote choices. In this article, I leverage a major educational reform to identify the effects of high school education on downstream voting behavior in Great Britain. In 1944, Winston Churchill’s cross- party coalition government pass legislation raising the high school leaving age from 14 to 15. The reform, which came into effect in 1947, induced almost half the student population to remain in school for either one or two additional years (but did not affect tertiary education progression). The magnitude of Britain’s 1947 reform marks it apart from leaving age reforms in North America and Europe (see Brunello, Fort and Weber 2009; Oreopoulos 2006), and experimental studies providing unrepresentative participants with incentives to remain in school (Sondheimer and Green 2010).

2 It therefore represents a unique opportunity to estimate education’s political effects for the lower half of the education distribution. I use a regression discontinuity (RD) design to compare voters from cohorts just old enough to be affected by the reform to voters from cohorts just young enough not to have been affected. I first identify the effect of the 1947 reform on the probability of voting for the Conservative party—Britain’s most prominent economically conservative party. Given that some students would have remained in school regardless of the higher leaving age, I then use the 1947 reform to instrument for schooling in order to identify the effect of an additional year of late high school for students that only remained in school because of the reform. The results show that staying in high school substantially increases the likelihood that an individual votes for the Conservative party in later life. In particular, I find that each additional year of high school increases the probability of voting Conservative by nearly 12 percentage points. For cohorts affected by the reform, this translates into a 4.4 percentage point increase in the Con- servative vote share. Although the RD estimates are local to the cohorts aged 14 around 1947, a correlation of similar magnitude holds between completing late high school and voting behavior across all cohorts. This supports the external validity of this finding, and suggests that completing high school is a key point at which education affects political preferences. Furthermore, the fact that a similar correlation holds away from the discontinuity implies that Britain’s 1947 reform changed the dynamics of national politics, and could have altered the outcomes of the close 1970 and 1992 Conservative election victories. Future educational reforms thus pose an important “catch 22” for the Labour and Liberal parties, who are ideologically committed to expanding educational opportunities for the least educated but also face an electoral cost of such policies. Beyond demonstrating that education causes voters to support the Conservative party, I provide evidence suggesting that education’s effects operate according to a Meltzer and Richard(1981) distributive logic. By increasing an individual’s income, education increases support for right-wing economic policies, which in turn leads individuals to vote Conservative. I provide a variety of evidence consistent with this mechanism. First, education significantly increases a student’s fu-

3 ture income (see Devereux and Hart 2010; Harmon and Walker 1995; Oreopoulos 2006), and only increases Conservative voting before retirement age. Second, and consistent with a permanent increase in income, an individual’s greater support for the Conservatives is relatively durable: an additional year of schooling causes individuals to self-identify as partisans, and increases the likelihood that they decided how they would vote before the start of the electoral campaign. Third, education increases support for economic policies associated with the Conservative party, including opposition to higher taxes, redistribution and welfare spending. Fourth, to demonstrate that educated individuals vote Conservative because of their policies, rather than the reverse relationship where voters simply adopt the positions of the party they identify with, I show that education does not affect support for non-economic positions associated with the Conservative party. Finally, I find no evidence that voters become more socially liberal or are impacted by more politically engaged social networks. This article proceeds as follows. I first consider the theoretical mechanisms potentially linking schooling and vote choice. I then describe Britain’s 1947 leaving age reform, the data and iden- tification strategy. The next section presents the main results. The penultimate section examines the mechanisms linking high school to voting Conservative. I then conclude by considering the implications of the results.

2 Why might high school education affect political preferences?

Arguably the most obvious channel through which education might affect political preferences is through an individual’s labor market position. An inﬂuential human capital literature argues that education imparts valuable skills that make workers more productive employees for ﬁrms (Becker 1964). These skills are generally rewarded in terms of higher wages (e.g. Angrist and Krueger 1991; Ashenfelter and Rouse 1998; Oreopoulos 2009). Linking education to political preferences, Romer(1975) and Meltzer and Richard(1981) (henceforth RMR) argue that workers

4 receiving higher wages will prefer lower income tax rates and lower government spending, particularly on means-tested programs, because they are net losers when tax revenues are progressively redistributed. Similar arguments may also apply to expected income, such that voters support conservative policies in anticipation of their higher future income (Alesina and La Ferrara 2005). In the British context, the human capital and RMR models imply that, by increasing their income, greater education should cause voters to become more favorable toward the Conservative party, and particularly the party’s relatively fiscally conservative platform. However, a more sociological literature has instead suggested that education cultivates socially liberal attitudes. Lipset(1959) famously proposed that education socializes liberal attitudes by directly communicating support for toleration and democracy. Hyman and Wright(1979) go further, arguing that—by expanding their frames of reference—education causes students to think in a fundamentally more liberal fashion. Furthermore, the final years of high school may also be a particularly important moment in the crystallization of lifelong political views (Ghitza and Gel- man 2014). In Britain, the Labour and Liberal Democrat parties are generally regarded as more socially progressive on issues of crime, immigration and giving voice to the disadvantaged. If education causes voters to become more socially liberal, then Labour and the Liberals may instead be expected to benefit electorally. Increasing a voter’s level of education may also expose them to new politically-relevant information and social networks in later life (e.g. Green, Palmquist and Schickler 2002; Nie, Junn and Stehlik-Barry 1996; Pattie and Johnston 2000). Although the extent of partisanship and political engagement in the social networks that educated voters might enter is not obvious, the least educated—the subjects of this study—may be differentially exposed to new conservative perspec- tives and information shortcuts that increase support for the Conservative party (Lupia 1994). Al- ternatively, educated voters could join politically engaged social networks, such as unions, that provide political information and social incentives to support the Labour or even the Liberal party (e.g. Abrams, Iversen and Soskice 2010).

5 Existing evidence examining the relationship between education and political preferences paints a mixed picture. On one hand, there is a robust survey-level correlation between individual income (which education increases) and opposition to taxation and redistribution across developed coun- tries (Alesina and La Ferrara 2005; Iversen and Soskice 2001; Shayo 2009). Furthermore, an individual’s income is positively correlated with support for right-wing parties in the United States (e.g. Gelman et al. 2010), the United Kingdom (e.g. Clarke et al. 2004; Whitten and Palmer 1996) and Western Europe (e.g. Thomassen 2005).1 On the other hand, however, the association between education and socially liberal attitudes and political engagement is also widely documented (e.g. Dee 2004; Nie, Junn and Stehlik-Barry 1996; Phelan et al. 1995). Rather than supporting right- wing parties, this impetus generally seems to push voters toward left-wing parties supporting more socially liberal policies (e.g. Heath et al. 1985; Inglehart 1981). However, it is hard to attribute a causal interpretation to these intriguing if seemingly conflict- ing associations. One major problem is that more educated individuals also differ in other key respects, such as possessing greater labor market potential (Spence 1973), coming from more af- fluent social backgrounds (Jencks et al. 1972), or being exposed to different social and political values as a child (Jennings, Stoker and Bowers 2009). In light of such concerns, Kam and Palmer (2008) suggest that education may simply “proxy” for other variables.2 Furthermore, interpreting existing estimates of education’s effects is problematic when most studies also control for various “post-treatment” variables—such as income, partisanship, and social networks—that are them- selves a function of education. Including such controls could induce severe post-treatment bias, and the direction of such bias is hard to establish (see King and Zeng 2007). This could explain why empirical analyses using different specifications yield very different conclusions. Experimental and quasi-experimental studies are required to disentangle the complex layers of 1However, the national-level implications of the RMR model have received mixed support (e.g. Karabarbounis 2011). 2Given education is closely tied with idiosyncratic experiences, it is unlikely that matching designs can resolve such problems (see Henderson and Chatfield 2011; Kam and Palmer 2008).

6 causality underpinning education’s political effects. Recent work using such methods has made signiﬁcant progress in identifying schooling’s effects on political participation (see Sondheimer and Green 2010). However, the external validity of such studies is often limited by focusing on a small and unrepresentative participant pool. Moreover, such methods have yet to be utilized to identify education’s effect on vote choice.

3 Empirical strategy

To estimate the effect of high school education on political preferences, I leverage Great Britain’s 1947 school leaving age reform as a source of exogenous variation. In particular, I use a RD design to compare essentially identical students born just too early and just early enough to be affected by the reform. Given the difﬁculty of identifying education’s political effects for a substantial proportion of the population, Britain’s 1947 reform—which affected nearly half the student population—represents a unique opportunity to disentangle the causal effects of education for a large and important segment of relatively uneducated voters.

3.1 Britain’s 1947 compulsory schooling reform

Britain’s education laws deﬁne the maximum age by which a student must start school and the minimum age at which they can leave school. In 1944, a time when barely 50% of students received any formal education beyond the age of 14, legislation was enacted to increase school leaving age from 14 to 15. The landmark reform was principally designed to create a fairer society in recognition of the population’s successful war effort, and was passed by Winston Churchill’s cross-party coalition government. The Education Act 1944 raised the leaving age in England and Wales, while the Education (Scotland) Act 1945 subsequently enacted the same reform in Scotland.3 The new leaving age, which had repeatedly failed to pass in the 1920s and 1930s due 3No such reform occurred in Northern Ireland until 1957, which is excluded from the analysis.

7 to financial constraints (Gillard 2011), came into force on April 1st 1947 after several years of intensive preparation. The Online Appendix describes the reform in greater detail, and locates it in the context of other (less major) educational reforms in Britain. The 1947 reform, which is arguably the largest post-war reform undertaken by any industrial- ized democracy, substantially increased educational attainment for a large proportion of Britain’s students. As shown in Figure1, the reform induced almost half of the student population to remain in school for at least an additional year. The majority only remained in school until age 15, but a non-trivial proportion continued until age 16 (the age at which most students complete high school). The proportion of students attending university, however, was unaffected. Therefore, in contrast to compulsory schooling reforms in Europe and North America that only affected a small and relatively unrepresentative set of students (see Brunello, Fort and Weber 2009; Oreopoulos 2006), Britain’s 1947 reform will allow me to identify the effect of increasing late high school education for almost the entire lower half of the education distribution. Given that the most significant post-war changes in the education system had already been implemented by 1947, the large rise in enrollment reflected the higher leaving age rather than other changes in the education system. Fees for secondary schooling were removed in 1944, while the new Tripartite system—which formally established three types of secondary school emphasizing academic, scientific and practical skills—came into force in 1945. However, as Figure1 indicates, these structural reforms did not affect enrollment (see also Oreopoulos 2006). Furthermore, prior to the 1947 reform, the government pre-emptively engaged in a major expansion effort to maintain school quality by increasing the number of teachers, buildings and classroom materials (Woodin, McCulloch and Cowan 2013). The additional year of schooling was primarily intended to ensure that students grasped the material they had previously been taught (Clark and Royer 2013).

8 r it-erchr vrgs n hi ierflcsterwih ntesample. the in weight their reflects size their and averages, cohort birth-year are Notes Proportion leaving 0 .2 .4 .6 .8 1 1925 aafo h E dsrbdblw.Cre ersn orhodrplnma t.Ge dots Grey fits. polynomial fourth-order represent Curves below). (described BES the from Data : iue1 97cmusr coln eomadsuetlaigaeb cohort by age leaving student and reform schooling compulsory 1947 1: Figure Leave before14 1930 1935 Leave before15 1940 Cohort: yearaged14 1945 Leave before16 9 1950 1955 Leave before17 1960 Leave before18 1965 1970 3.2 Data

I use data from the British Election Survey (BES) to examine the reform’s political implications. The BES, which randomly samples voting age citizens with British postal addresses for in-person interviews,4 has been conducted following every general election since 1964. Using the nine elections from 1974 to 2010, where the relevant variables are available, produced a maximum sample of 24,439 observations. The empirical analysis utilizes three key variables. First, the principal outcome is an indicator for voting Conservative at the last election. In the sample, 35% of respondents reported voting Conservative, while 37% and 19% respectively reported voting Labour and Liberal. Suggesting that reported voting relatively reliable, the survey-weighted Conservative vote share across period under study is 37%. To understand how changes in Conservative support affect other parties, I will also examine indicators for voting Labour and Liberal. Second, I deﬁne the minimum schooling leaving age for an individual in (birth year) cohort c by an indicator for whether the reform was

5 binding when the student was aged 14, i.e. Post 1947 reformc = 1(Birth year + 14 ≥ 1947). Fi- nally, I measure education as the number of years of schooling. This was computed by subtracting five—the age at which students start school—from the age at which a respondent reported leaving formal schooling. Given that using a binary measure of an endogenous treatment variable such as completing high school can substantially upwardly bias IV estimate (Marshall 2015), years of schooling represents a conservative coding approach that guarantees a consistent estimate of the average effect of an additional year of schooling. The Online Appendix provides detailed variable definitions and summary statistics. 4Additional pre-election and non-interview surveys were excluded. 5Month of birth is unavailable, so the instruments are assigned by birth year. However, my first stage is very similar to Clark and Royer(2013), who can assign the instruments using month of birth data. The clear graphical discontinuity further supports this coding.

10 3.3 Identiﬁcation and estimation

To identify the effect of late high school education on vote choice, I exploit Britain’s 1947 school leaving age reform as a natural experiment. Among cohorts aged around 14 in 1947, being subject to the higher leaving age effectively randomly assigned a strong incentive to remain in school for an additional year. Accordingly, I employ a RD design to identify the effect of the reform, where the running variable determining whether an individual is “treated” by the 1947 reform is an individual’s birth year cohort. Since the reform could not force every student to remain in school, to estimate the effect of an additional year of late high school education I also exploit a “fuzzy” RD design where the 1947 reform is used as an instrument discontinuously increasing the probability of receiving an additional year of education (see Hahn, Todd and Van der Klaauw 2001).6 The key identifying assumption is that the decision to vote Conservative is continuous across cohorts at the reform discontinuity in all variables other than the school leaving age. In this particular case, there are good reasons to doubt the “sorting” concern that another key variable simultaneously changes at the discontinuity. First, selection into cohorts in Britain is implausible since parents could not have precisely predicted the 1947 reform more than a decade in advance. McCrary(2008) tests in the Online Appendix confirm that there is no discontinuous change in the mass of respondents in the sample that were born either side of the reform. Second, broad shifts in political culture are unlikely to have affected 15 year olds without also affecting 14 year olds. Flex- ible birth year trends are also included to address this concern. Furthermore, since cohorts born either side of the cutoff were first eligible to vote at the 1955 election, there is no differential “first election” effect impacting students facing a higher leaving age (e.g. Meredith 2009; Mullainathan and Washington 2009). Third, Figure2 shows that pre-treatment demographic, socio-economic and labor market characteristics are essentially continuous through the discontinuity.7 Fourth, the 6Given the dramatic change in educational attainment it induced across neighboring cohorts, the reform has proved popular as an instrument for education among labor economists (see Clark and Royer 2013; Oreopoulos 2006). However, it has not been used in a political context. 7Tests in the Online Appendix confirm that there is no significant change in the gender or racial

11 eue omrgeso sn OLS: using regression form reduced changes. institutional or social 1947 proximate the other that or suggest pre-trends tests capturing placebo simply These not reforms. is placebo reform as 1947 to treat- prior when years support ten Conservative the of in any change ing signiﬁcant no is there that indicates Appendix Online opsto ftesml rtepooto frsodnswoefteswr aulworkers manual were fathers whose respondents of reform. proportion 1947 the the or around sample the of composition Panel G:Nationalunemploymentrate iue2 rnsi eorpi,scoeooi n ao aktdmgahcvariables demographic market labor and socio-economic demographic, in Trends 2: Figure oetmt h feto h 97rfr tefo oigCnevtv,Ietmt h following the estimate I Conservative, voting on itself reform 1947 the of effect the estimate To

nln U cnmcDt 7020”dataset. 1700-2009” Data Economic “UK England Notes Rate (%) Proportion Year 0 5 10 15 0 .01 .02 1980 2000 1930 h aai aesAFaefo h E.Tedt nPnl n sfo h akof Bank the from is H and G Panels in data The BES. the from are A-F Panels in data The : 1930 1930 Panel A:Surveyyear Cohort: yearaged14 Cohort: yearaged14 Cohort: yearaged14 oeConservative Vote 1940 Panel D:Black 1940 1940 1950 1950 1950 1960 1960 1960 1970 1970 1970 ic = Index (2000=100) δ Panel H:Nationalaverageearnings Proportion Proportion

ot14 reform 1947 Post 0 2 4 6 0 .01 .02 .03 .4 .45 .5 .55 1930 1930 1930 Cohort: yearaged14 Cohort: yearaged14 Cohort: yearaged14 1940 Panel E:Asian Panel B:Male 1940 1940 12 1950 1950 1950 c 1960 1960 1960 + f ( 1970 1970 1970 it year Birth

Proportion Panel F:Fathermanual/unskilledjob Proportion .6 .7 .8 .96 .98 1 c + ) 1930 1930 Cohort: yearaged14 Cohort: yearaged14 ε ic 1940 Panel C:White 1940 (1) , 1950 1950 1960 1960 1970 1970 where f is a function of the running variable used to control for trends in Conservative support away from the discontinuity. In particular, I estimate local linear regressions (LLRs) where f includes linear birth year trends either side of the discontinuity, and only observations within the Imbens and Kalyanaraman(2012) optimal bandwidth (of 14.7 cohorts) are included in the sample. The average age of a respondent in this sample is 56. To ensure the comparability of treated and untreated cohorts, observations are weighted by their proximity to the discontinuity using a triangle kernel.8 As robustness checks, I show below that the results do not depend upon the choice of bandwidth, kernel, or polynomial order of the cohort trends. The principal theoretical quantity of interest, however, is the effect of schooling on voting Conservative. To estimate this, I instrument for years of schooling using the 1947 reform. Beyond the standard RD assumption discussed above, identiﬁcation of schooling’s effect on Conserva- tive voting also requires that the instrument (a) never decreases an individual’s level of education (monotonicity) and (b) only affects voting through education (exclusion restriction). As show in Figure1, and consistent with monotonicity, the proportion of students leaving school at any age never increases. Given the proximity of the reform to the choice to remain in school, it is unlikely that raising the leaving age affected an individual’s political preferences through channels other than additional schooling. Nevertheless, potential violations of the exclusion restriction are discussed below. To estimate the effects of an additional year of schooling among respondents that only remained in school because of the reform, I estimate the following structural equation using 2SLS:

Vote Conservativeic = βSchoolingic + f (Birth yearc) + εic, (2) 8Estimation uses the Stata command rdrobust (Calonico, Cattaneo and Titiunik 2014).

13 where exogenous variation in schooling comes from the the ﬁrst stage regression below:

Schoolingic = αPost 1947 reformc + f (Birth yearc) + εic. (3)

The results demonstrate that the strength of the ﬁrst stage far exceeds the F statistic of 10 required to safely dismiss weak instrument bias (Staiger and Stock 1997).

4 High school education’s effect on vote choice

This section presents the paper’s main result: high school education causes a substantial increase in support for the Conservative party later in life. I ﬁrst present the effect of the 1947 on years of schooling and downstream support for Conservatives, before turning to the instrumental variable estimates identifying the effect of an additional year of late high school.

4.1 Britain’s 1947 reform increases schooling and Conservative voting

Confirming the dramatic increase in schooling registered in Figure1, the first stage estimate in column (1) of Table1 shows that the 1947 reform substantially increased education attainment. Specifically, the reform increased the schooling of an average student by 0.38 years. The F statistic of 25.4 indicates a strong first stage. Turning to Conservative voting, the reduced form plot in Figure3 provides the first evidence that voters either side of the reform differ systematically in their vote choice. In particular, there is a notable jump in Conservative voting among cohorts affected by the 1947 reform. The fact that increasing the school leaving age reverses the relatively secular trend against the Conservatives— which is likely to be a function of both declining support over time and younger voters being more left-wing—adds weight to the plausibility of the relationship by suggesting that it does not simply reflect accelerating cohort trends.

14 Table 1: Estimates of schooling’s effect on voting Conservative

Years Vote Vote Vote Vote Vote Vote of Con. Con. Con. Con. Labour Liberal schooling LLR LLR LLR IV OLS OLS LLR IV LLR IV (1) (2) (3) (4) (5) (6) (7) Post 1947 reform 0.381*** 0.044** (0.076) (0.020) Years of schooling 0.116** 0.021*** -0.071 -0.021 (0.056) (0.002) (0.052) (0.043) 8th year of schooling -0.020 (0.036) 9th year of schooling Baseline

10th year of schooling 0.126*** (0.013) 11th year of schooling 0.213*** (0.014) 12th year of schooling 0.289*** (0.017) 13th year of schooling 0.306*** (0.018) 14th year of schooling 0.281*** (0.020)

Observations 11,068 11,068 11,068 16,757 16,757 11,068 11,068 First stage F statistic 25.4 25.4 25.4 25.4

Notes: Specification (1) is the first stage estimates of the 1947 reform’s effect on years of schooling. Specification (2) is the reduced form estimate of the 1947 reform on Conservative voting. Specification (3) is the IV estimate for years of schooling. All specifications, excluding (4), are local linear regressions using a triangular kernel and the Imbens and Kalyanaraman(2012) optimal bandwidth of 14.7. Specifications (4) and (5) are OLS regressions (in the full BES sample) of voting Conservative on years of schooling (separately as a continuous variable and a set of indicators for each year of schooling), controlling for indicators for male, black, white and Asian respondents, standardized quartic polynomials in age and birth year cohort, and survey fixed effects. For the set of schooling indicators, the estimates for other years are omitted from this table. Robust standard errors in parentheses. * denotes p < 0.1, ** denotes p < 0.05, *** denotes p < 0.01.

15 osaebrhya ootaeae,adtersz eet hi egti h sample. the in weight their reflects size their and averages, cohort birth-year are dots Notes Conservative vote share .2 .3 .4 .5 1925 lc uvsrpeetfut-re oyoilfisete ieo h 97dsotniy Grey discontinuity. 1947 the of side either fits polynomial fourth-order represent curves Black : 1930 iue3 rprinvtn osraieb it ercohort year birth by Conservative voting Proportion 3: Figure 1935 1940 Cohort: yearaged14 1945 16 1950 1955 1960 1965 1970 More formally, column (2) of Table1 estimates the reduced form effect of the reform on voting Conservative later in life. The coefficient indicates that increasing the leaving age to 15 induced a large and statistically significant increase in support for the Conservative party. Students from each cohort affected by the 1947 reform are 4.4 percentage points more likely to vote Conservative. Relative to 35% of the sample that vote Conservative, this implies that affected cohorts are around 13% more Conservative. This large difference implies that the reform substantially altered national politics, and could easily have altered the outcomes of the close Conservative election victories in 1970 and 1992. If the effects at the discontinuity generalize to more recent cohorts where completing high school education is the norm, the reform’s legacy becomes increasingly important as the proportion of pre-reform voters in the population declines.

4.2 High school education’s increases Conservative voting

By averaging across all individuals in a given cohort, and thus including students that would have remained in school regardless of the reform, the reduced form underestimates the 1947 reform’s impact on individuals who only remained in school because the leaving age increased. To calculate the effects for such compliers, I turn to the IV/fuzzy RD estimates. Instrumenting for years of schooling, column (3) presents the average effect of an additional year of schooling for compliers. Late high school substantially increases the probability of voting Conservative later in life—in fact, each additional year of high school increases this probability by almost 12 percentage points. Reinforcing the reduced form estimates—and consistent with surveys documenting a positive correlation between voting Conservative and greater education and higher social class (e.g. Clarke et al. 2004; Whitten and Palmer 1996)—this large and statistically significant coefficient provides clear causal evidence that high school education is a major determinant of long-run conservative political behavior among the least educated. This finding most obviously fits with the income-based channels considered above, although further evidence supporting this mechanism is presented below.

17 By way of comparison, column (4) estimates the correlation between years of schooling and Conservative voting in the full sample. Including controls for gender and race, as well as flexible polynomials in age and cohort, the estimates suggest that each additional year of schooling is associated with a 2 percentage point greater likelihood of voting Conservative. However, this average masks important features of the correlation between education and vote choice. Column (5) shows that even in the full BES sample, the coefficients for the 10th and 11th years of schooling—which generally correspond to leaving school at ages 15 and 16—are similar in magnitude to the IV estimates. Although there are insufficient instruments to estimate such a non-linear effect in the IV context, this suggests that late high school is a particularly consequential moment in a adolescent’s life trajectory. The significant drop off in the correlation after high school also offers tentative support for the possibility that education’s political effects are not linear. However, it is important to reiterate that because the reform did not increase university attendance the causal estimates exploiting the 1947 reform are local to high school education and cannot identify whether university similarly affects voting behavior. In Britain’s three-party system, it is not obvious which party loses potential supporters to the Conservatives. To address this question, columns (6) and (7) respectively estimate the effect of schooling on voting for the Labour and Liberal parties. Although neither coefficient is precisely estimated, the results suggest that Labour are the principal losers: an additional year of high school education decreases the probability of voting Labour by 7 percentage points, whereas the Liberals only suffer a 2 percentage point decline. Given that surveys typically document greater Liberal support among better-educated respondents (e.g. Sanders 2003), this smaller decline is relatively unsurprising. Nevertheless, the fact that greater education did not boost support for the Liberals suggests that the commonly-cited association between education and support for the Liberals may reflect other characteristics of educated voters, or may only arise from university education.

18 4.3 Robustness checks

I now demonstrate the robustness of the results to various potential concerns. First, the results are not artefacts of the particular RD specification used for the main estimates. Figure4 shows that the point estimates are stable across bandwidths and the choice of (triangular or rectangular) kernel. Inevitably, the precision of the estimates declines at the smaller bandwidths with fewer observations. Although the point estimates are remarkably stable across bandwidths, I also adjust for potential biases that could arise from selecting an optimal bandwidth that trades off bias against the efficiency gained from including observations further from the discontinuity. Correcting for such bias using the approach proposed by Calonico, Cattaneo and Titiunik(2014), the estimates (in the Online Appendix) are almost identical, and thus reinforce the robustness of the finding with respect to bandwidth. Furthermore, to demonstrate that the results are not being driven by complex trends across cohorts, the Online Appendix shows similar estimates when using higher- order polynomial cohort trends and finds no significant change around placebo reforms at any of the ten prior years. Second, the exclusion restriction (required for the IV estimates) is violated if the 1947 reform affected political preferences through channels other than schooling. Although political or cul- tural changes are unlikely to have differentially affect cohorts one year apart, it is possible that an additional year in school could affect life choices—such as marriage or having children—by simply keeping students in school, but without operating through schooling itself. To address such concerns, I examine these possibilities using Labor Force Surveys from the same years as the BES data. The Online Appendix shows that the 1947 reform did not affect the age of a respondent’s oldest (dependent) child, the number of children a respondent has, or whether the respondent has ever been married at the time of the survey. Furthermore, any reduction in schooling quality or spillover causing older cohorts to behave more like treated cohorts would reduce between-cohort differences around the reforms, and thus downwardly bias the estimates.

19 ersn 5 ofiec nevl frrbs tnaderrors). standard robust (for intervals confidence 95% represent Notes -.2 0 .2 .4 -.1 -.05 0 .05 .1 .15 2 2 rageadrcaglrdnt h hieo enlue o h pcfiain nec lt Bars plot. each in specifications the for used kernel of choice the denote rectangular and Triangle : 3 3 Effect ofyearsschooling(triangle) 4 4 Effect of1947reform(triangle) 5 5 6 6 7 7 8 8 iue4 outest hieo adit n kernel and bandwidth of choice to Robustness 4: Figure 9 9 Bandwidth Bandwidth 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 20

-.2 0 .2 .4 -.1 -.05 0 .05 .1 .15 2 2 Effect ofyearsschooling(rectangular) 3 3 Effect of1947reform(rectangular) 4 4 5 5 6 6 7 7 8 8 9 9 Bandwidth Bandwidth 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 5 How does schooling affect vote choice?

To illuminate the mechanisms causing late high school education to substantially increase downstream Conservative voting, I leverage additional questions from the BES surveys, placebo tests and heterogeneous effects. Although demonstrating a causal mechanism is difﬁcult, examining a range of potential mediators in conjunction with placebo tests can support some mechanisms and eliminate others (Gerber and Green 2012). These results principally suggest that education increases income, which in turn increases support for right-wing policies, and ultimately induces an individual to vote Conservative.

5.1 Greater income and persistent Conservative voting

The combination of human capital theory and the RMR model of income-based political preferences predict that education induces more conservative fiscal policy preferences by increasing an individual’s income. There exists compelling evidence that the 1947 reform increased the income of affected cohorts. Exploiting similar RD designs using Britain’s 1947 reform, previous studies have estimated that an additional year of schooling increases wage income by 5-15 percent (De- vereux and Hart 2010; Harmon and Walker 1995; Oreopoulos 2006). This significant increase in annual income over the course of a working life has the potential to alter political behavior. If education is indeed driving support for the Conservative party by increasing an individual’s income, education’s should predominantly affect respondents in the workforce. Once retired, education’s ability to generate higher wages may no longer be relevant. To test whether schooling ceases to affect vote choice once a respondent retires from the labor market, I compare the estimates for schooling between respondents aged above and below 60 years of age.9 Using specifications 9Since current employment may be endogenous to schooling, I use an age-based cutoff. Al- though workers increasingly retire in their 60s, the cutoff is chosen to conservatively capture retired respondents, and by including respondents still in the workforce should if anything under-estimate the difference. However, I find very similar results when 65 and 70 are used as cutoffs.

21 Table 2: Schooling, Conservative voting and income-based mechanisms

Vote Con. Vote Con. Con. Decided (below 60) (60 or above) partisan before campaign (1) (2) (3) (4) Panel A: Reduced form estimates Post 1947 reform 0.049** 0.006 0.039* 0.041** (0.024) (0.029) (0.022) (0.019)

Panel B: IV estimates Years of schooling 0.111** 0.027 0.092* 0.103* (0.056) (0.138) (0.052) (0.051)

Observations 10,252 4,589 9,711 9,510 Bandwidth 22.6 15.0 13.9 12.5 First stage F statistic 28.3 2.7 26.4 23.4

Notes: All specifications are local linear regressions using a triangular kernel and the Imbens and Kalyanaraman (2012) optimal bandwidth for each sample. Robust standard errors in parentheses. * denotes p < 0.1, ** denotes p < 0.05, *** denotes p < 0.01. analogous to equation (2), the results in Table2 support this implication. The reduced form and IV estimates in column (1) show that respondents aged below 60 experience large increases in their probability of voting Conservative commensurate to the estimates in Table1. However, consistent with schooling’s conservative effects only operating among active workers earning an income, the reduced form estimate in column (2) indicates that elderly respondents affected by the 1947 reform are no more likely to vote Conservative. Even with a weaker first stage, the IV estimate in column (2) is more than one quarter smaller than the effect of schooling among working-age respondents. These results suggest that education only affects vote choice to the extent that workers are continuing to accrue higher wages because of their greater education. Furthermore, since education’s economic returns are likely to hold throughout an individual’s working life, the decision to support the Conservative party should be relatively durable. Table2 also provides support for this claim. First, column (3) shows that education significantly increases

22 the likelihood that an individual self-identifies as a Conservative partisan. Since partisanship likely entails a deeper and more persistent attachment than just voting for a party at the previous election (e.g. Campbell et al. 1960; Clarke and Stewart 1998), the results imply that education forges a lasting tie with the Conservative party. Second, the finding in column (4) that educated voters are significantly more likely to decide how they will vote before the electoral campaign starts further suggests that, consistent with a permanent increase in income, schooling durably increases Conservative support.

5.2 Education increases support for Conservative economic policies

Given that education increases income, education should then also increase support for conservative economic policies such as lower taxation and lower redistributive spending (Meltzer and Richard 1981). To test this implication, I examine how education affects three economic policy attitudes: opposition to tax and spend policies, opposition to income and wealth redistribution, and the belief that welfare spending has gone too far. Columns (1)-(3) in Table3 suggest that increased education translates into more right-wing fiscal policy preferences. For each variable, the reduced form and IV estimates are large and positive, and only fail to reach statistical significant in the case of redistribution. Furthermore, combining these variables as a simple (standardized) additive scale (Cronbach’s alpha of 0.42), column (4) shows that an additional year of late high school increases support for conservative economic policies by 0.3 standard deviations. Although the causal link from fiscal policy preferences to vote choice cannot be causally identified, there is a strong negative correlation between voting Conservative and opposing high taxation, redistribution and welfare spending.10 However, if voters adopt the policy positions of the political party or candidate they identify with (e.g. Lenz 2012; Zaller 1992), changes in economic policy preferences could reflect changes 10The significant correlations between voting Conservative and opposing tax and spend, not supporting welfare benefits, and opposing redistribution are respectively 0.25, 0.41 and 0.21.

23 Table 3: Mechanisms through which schooling affects political preferences

Oppose Welfare Oppose Con. Support Support Oppose Political Union tax and beneﬁts redist. economic crime leaving abolishing information member spend gone policy reduction Europe private index too far scale (over rights) education (1) (2) (3) (4) (5) (6) (7) (8) (9) Panel A: Reduced form estimates Post 1947 reform 0.229* 0.056** 0.081 0.121*** 0.007 -0.007 0.024 0.054 0.009 (0.120) (0.025) (0.060) (0.040) (0.168) (0.023) (0.023) (0.054) (0.022)

24 Panel B: IV estimates Years of schooling 0.567* 0.100** 0.149 0.290*** 0.016 -0.017 0.035 0.140 0.020 (0.317) (0.047) (0.109) (0.104) (0.373) (0.053) (0.035) (0.133) (0.045)

Outcome range 0 to 10 0 or 1 0 to 4 -1.9 to 3.5 0 to 10 0 or 1 0 or 1 -5.2 to 1.7 0 or 1 Outcome mean 3.35 0.27 1.66 0.00 6.51 0.32 0.19 0.00 0.25 Outcome standard deviation 2.58 0.44 1.22 1.00 2.71 0.47 0.39 1.00 0.43 Bandwidth 14.6 17.7 16.7 18.0 18.9 13.7 14.1 14.1 10.9 Observations 7,370 6,928 8,231 11,793 5,716 8,561 5,610 6,097 7,354 First stage F statistic 18.5 31.4 30.7 31.0 21.2 22.2 25.3 11.8 23.2

Notes: For all outcomes, larger values are more pro-Conservative views. All specifications are local linear regressions using a triangular kernel and the variable’s respective optimal bandwidth (given the number of observations per variable differs). Robust standard errors in parentheses. * denotes p < 0.1, ** denotes p < 0.05, *** denotes p < 0.01. in partisanship rather than income-based incentives. To test this possibility, I examine whether respondents also adopt Conservative positions on three non-economic issues: emphasis on reducing crime over protecting citizen rights, support for Britain leaving the European community (EEC, EC or EU, depending on the survey year), and not abolishing private education.11 The results of these placebo tests—in columns (5)-(7)—show that education does not significantly shift voters toward any of these Conservative positions. These results, in addition to finding that education increases support for the Conservative party, further indicate that high school education does not increase support for socially liberal values. The evidence thus suggests that education’s political effects operate through fiscal policy preferences. As noted above, another explanation for education increasing support for the Conservative party is that individuals enter more politically engaged social networks. To examine this possibility, I examine a political information index which standardizes the proportion of factual political questions a respondent correctly answers across surveys, and an indicator of union membership. The results in columns (8) and (9) do not indicate that voters are either significantly more politically informed or less likely to be a member of a union.12 Of course, such measures cannot definitively rule out a social networks explanation. However, in combination with the finding that only economic policy preferences are altered, there is clearly no support for a network explanation where voters simply join generally right-wing groups that increase their political engagement.

6 Conclusion

In this article, I have shown that an additional year of late high school education substantially increases the probability that an individual votes for the Conservative party later in life. By lever- 11Unsurprisingly, emphasizing crime reduction (ρ = 0.12), not abolishing private education (ρ = 0.26) and leaving Europe (ρ = 0.03) are signiﬁcantly positively correlated with voting Con- servative. 12Unreported results also show that there is no change in turnout.

25 aging a RD design to exploit exogenous variation in the cohorts affected by a major reform that increased the education of almost half the British population, I am able to overcome the identifi- cation challenges encountered by previous studies to estimate the causal effect of late high school for a significant group of students that only stayed in school because of the landmark reform. Furthermore, I provide evidence suggesting that education’s large effects operate by increasing income, which increases support for right-wing economic policies and ultimately support for the Conservative party. These findings suggest that by affecting an individual’s labor market prospects and associated political preferences, high school education may be one of the most important determinants of voting behavior. The evidence thus supports influential studies arguing that early life events have important downstream consequences (see Jennings, Stoker and Bowers 2009; Kam and Palmer 2008). However, the mechanism is not the liberal socialization claim frequently attributed to schooling (Hyman and Wright 1979; Lipset 1959). Rather, high school education’s effects appear to be predominantly economic in nature and most salient among active members of the workforce, and thus consistent with the contested distributive logic of Meltzer and Richard(1981). One possible explanation for previous correlational studies attributing liberal effects to education is that education only appear to have liberal effects once post-treatment measures of income are also controlled for. Furthermore, the effects of Britain’s 1947 leaving age reform are sufficiently large to have potentially affected subsequent electoral outcomes. Affected cohorts are nearly 5 percentage points more likely to vote for the Conservative party, which could have won the 1970 and 1992 elections because of the reform. To the extent that the reform has altered the composition of political office, it could have meaningfully altered policy outcomes (e.g. Lee, Moretti and Butler 2004). As noted above, these findings pose a strategic dilemma for the Labour and Liberal parties, which have typically supported progressive education policies providing opportunities for the least educated. Given that RD designs can only estimate causal effects at a given discontinuity, it is important to consider the generality of the results beyond the impact of schooling around the 1947 reform.

26 First, it should be reiterated that unlike many quasi-experimental methods, Britain’s 1947 reform affect almost half the student population, and thus speaks to an extensive set of voters affected by the reform. Second, although it is only suggestive, the RD point estimates are very similar to those in the full BES sample covering all available cohorts. Third, the full sample correlations also suggest that education’s political effects may peak at high school. The impact of further education appears to be relatively minimal, although future research is required to estimate such effects with more recent data. As the economic returns to education have changed, it is possible that university education has become the key driver of education’s political effects. Alternatively, education’s income effects are perhaps offset by socially liberal attitudes only instilled at university.

27 References

Abrams, Samuel, Torben Iversen and David Soskice. 2010. “Informal Social Networks and Ratio- nal Voting.” British Journal of Political Science 41:229–257.

Alesina, Alberto and Eliana La Ferrara. 2005. “Preferences for Redistribution in the Land of Opportunities.” Journal of Public Economics 89(5):897–931.

Angrist, Joshua D. and Alan B. Krueger. 1991. “Does Compulsory School Attendance Affect Schooling and Earnings?” Quarterly Journal of Economics 106(4):979–1014.

Ashenfelter, Orley and Cecilia Rouse. 1998. “Income, schooling, and ability: Evidence from a new sample of identical twins.” Quarterly Journal of Economics 113(1):253–284.

Becker, Gary S. 1964. Human Capital: A Theoretical and Empirical Analysis, with Special Refer- ence to Education. University of Chicago Press.

Brunello, Giorgio, Margherita Fort and Guglielmo Weber. 2009. “Changes in compulsory schooling, education and the distribution of wages in Europe.” Economic Journal 119(536):516–539.

Calonico, Sebastian,´ Mat´ıas Cattaneo and Roc´ıo Titiunik. 2014. “Manipulation of the running variable in the regression discontinuity design: A density test.” Econometrica 82(6):2295–2326.

Campbell, Angus, Philip E. Converse, Warren E. Miller and Donald E. Stokes. 1960. The American Voter. New York: Wiley.

Clark, Damon and Heather Royer. 2013. “The Effect of Education on Adult Mortality and Health: Evidence from Britain.” American Economic Review 103(6):2087–2120.

Clarke, Harold D., David Sanders, Marianne C. Stewart and Paul Whiteley. 2004. Political Choice in Britain. Oxford University Press.

28 Clarke, Harold D. and Marianne C. Stewart. 1998. “The decline of parties in the minds of citizens.” Annual Review of Political Science 1(1):357–378.

Converse, Philip E. 1972. Change in the American Electorate. In The Human Meaning of Social Change, ed. Angus Campbell and Philip E. Converse. New York, NY: Russell Sage Foundation pp. 263–337.

Dee, Thomas S. 2004. “Are there civic returns to education?” Journal of Public Economics 88:1697–1720.

Devereux, Paul J. and Robert A. Hart. 2010. “Forced to be Rich? Returns to Compulsory Schooling in Britain.” Economic Journal 120:1345–1364.

Gelman, Andrew, Park, Boris Shor, Joseph Bafumi and Jeronimo Cortina. 2010. Red State, Blue State, Rich State, Poor State: Why Americans Vote the Way They Do. Princeton, NJ: Princeton University Press.

Gerber, Alan S. and Donald P. Green. 2012. Field Experiments: Design, Analysis, and Interpreta- tion. W.W. Norton.

Gerber, Alan S., Gregory A. Huber, David Doherty, Conor M. Dowling and Shang E. Ha. 2010. “Personality and Political Attitudes: Relationships Across Issue Domains and Political Con- texts.” American Political Science Review 104(01):111–133.

Ghitza, Yair and Andrew Gelman. 2014. “The Great Society, Reagan’s Revolution, and Genera- tions of Presidential Voting.” Working paper.

Gillard, Derek. 2011. “Education in England: A Brief History.” Web link.

Green, Donald P., Bradley Palmquist and Eric Schickler. 2002. Partisan Hearts and Minds: Polit- ical Parties and the Social Identities of Voters. Yale University Press.

29 Hahn, Jinyong, Petra Todd and Wilbert Van der Klaauw. 2001. “Identiﬁcation and estimation of treatment effects with a regression-discontinuity design.” Econometrica 69(1):201–209.

Harmon, Colm and Ian Walker. 1995. “Estimates of the Economic Return to Schooling for the United Kingdom.” American Economic Review 85(5):1278–1286.

Heath, Anthony, Roger Jowell, John Curtice, Julia Field and Clarissa Levine. 1985. How Britain Votes. Pergamon Press Oxford.

Henderson, John and Sara Chatﬁeld. 2011. “Who Matches? Propensity Scores and Bias in the Causal Effects of Education on Participation.” Journal of Politics 73(3):646–658.

Hyman, Herbert H. and Charles R. Wright. 1979. Education’s Lasting Inﬂuence on Values. Uni- versity of Chicago Press.

Imbens, Guido W. and Karthik Kalyanaraman. 2012. “Optimal bandwidth choice for the regression discontinuity estimator.” Review of Economic Studies 79(3):933–959.

Inglehart, Ronald. 1981. “Post-Materialism in an Environment of Insecurity.” American Political Science Review 75(4):880–900.

Iversen, Torben and David Soskice. 2001. “An Asset Theory of Social Policy Preferences.” Amer- ican Political Science Review 95(4):875–894.

Jencks, Christopher, Marshall Smith, Henry Acland and Mary Jo Bane. 1972. Inequality: A Re- assessment of the Effect of Family and Schooling in America. New York, NY: Basic Books.

Jennings, M. Kent, Laura Stoker and Jake Bowers. 2009. “Politics across generations: Family transmission reexamined.” Journal of Politics 71(3):782–799.

Kam, Cindy D. and Carl L. Palmer. 2008. “Reconsidering the Effects of Education on Political Participation.” Journal of Politics 70(3):612–631.

30 Karabarbounis, Loukas. 2011. “One Dollar, One Vote.” Economic Journal 121(553):621–651.

King, Gary and Langche Zeng. 2007. “When can history be our guide? the pitfalls of counterfac- tual inference1.” International Studies Quarterly 51(1):183–210.

Lee, David S., Enrico Moretti and Matthew J. Butler. 2004. “Do voters affect or elect policies? Evidence from the US House.” Quarterly Journal of Economics pp. 807–859.

Lenz, Gabriel S. 2012. Follow the Leader? How Voters Respond to Politicians’ Policies and Performance. University of Chicago Press.

Lipset, Seymour Martin. 1959. “Some social requisites of democracy: Economic development and political legitimacy.” American political science review 53(1):69–105.

Lupia, Arthur. 1994. “Shortcuts Versus Encyclopedias: Information and Voting Behavior in Cali- fornia Insurance Reform Elections.” American Political Science Review 88(1):63–76.

Marshall, John. 2015. “Coarsening bias: How instrumenting for coarsened treatments upwardly biases instrumental variable estimates.” Working paper.

McCrary, Justin. 2008. “Manipulation of the running variable in the regression discontinuity design: A density test.” Journal of Econometrics 142(2):698–714.

Meltzer, Allan H. and Scott F. Richard. 1981. “A rational theory of the size of government.” Journal of Political Economy 89:914–927.

Meredith, Marc. 2009. “Persistence in Political Participation.” Quarterly Journal of Political Sci- ence 4(3):187–209.

Milligan, Kevin, Enrico Moretti and Philip Oreopoulos. 2004. “Does education improve citizen- ship? Evidence from the United States and the United Kingdom.” Journal of Public Economics 88:1667–1695.

31 Mullainathan, Sendhil and Ebonya Washington. 2009. “Sticking with Your Vote: Cognitive Disso- nance and Political Attitudes.” American Economic Journal: Applied Economics 1(1):86–111.

Nie, Norman H., Jane Junn and Kenneth Stehlik-Barry. 1996. Education and Democratic Citizen- ship in America. University of Chicago Press.

Ofﬁce for National Statistics. 2013. “Vital Statistics: Population and Health Reference Tables— Annual Time Series Data.” Web link.

Oreopoulos, Philip. 2006. “Estimating Average and Local Average Treatment Effects of Education when Compulsory Schooling Laws Really Matter.” American Economic Review 96(1):152–175.

Oreopoulos, Philip. 2009. Would More Compulsory Schooling Help Disadvantaged Youth? Evi- dence from Recent Changes to School-Leaving Laws. In The Problems of Disadvantaged Youth: An Economic Perspective, ed. Jonathan Gruber. University of Chicago Press pp. 85–112.

Pattie, Charles and Ron Johnston. 2000. “People who talk together vote together: An exploration of contextual effects in Great Britain.” Annals of the Association of American Geographers 90(1):41–66.

Phelan, Jo, Bruce G. Link, Ann Stueve and Robert E. Moore. 1995. “Education, social liberalism, and economic conservatism: Attitudes toward homeless people.” American Sociological Review 60(1):126–140.

Romer, Thomas. 1975. “Individual welfare, majority voting, and the properties of a linear income tax.” Journal of Public Economics 4(2):163–185.

Sanders, David. 2003. “Party identiﬁcation, economic perceptions, and voting in British general elections, 1974–97.” Electoral Studies 22(2):239–263.

Shayo, Moses. 2009. “A model of social identity with an application to political economy: Nation, class, and redistribution.” American Political Science Review 103(2):147–174.

32 Sondheimer, Rachel M. and Donald P. Green. 2010. “Using Experiments to Estimate the Effects of Education on Voter Turnout.” American Journal of Political Science 41(1):178–189.

Spence, Michael. 1973. “Job market signaling.” Quarterly Journal of Economics 87(3):355–374.

Staiger, Douglas and James H. Stock. 1997. “Instrumental Variables Regression with Weak Instru- ments.” Econometrica 65(3):557–586.

Thomassen, Jacques J.A. 2005. The European Voter: A Comparative Study of Modern Democra- cies. Oxford University Press.

Whitten, Guy D. and Harvey D. Palmer. 1996. “Heightening comparativists’ concern for model choice: voting behavior in Great Britain and the Netherlands.” American Journal of Political Science pp. 231–260.

Woodin, Tom, Gary McCulloch and Steven Cowan. 2013. “Raising the participation age in historical perspective: policy learning from the past?” British Educational Research Journal 39(4):635–653.

Woodin, Tom, McCulloch Gary and Steven Cowan. 2013. Secondary Education and the Raising of the School Leaving Age: Coming of Age? New York: Palgrave MacMillan.

Zaller, John. 1992. The Nature and Origins of Mass Opinion. Cambridge University Press.

33 Appendix

Brief history of British education reforms in the twentieth century

There have been three landmark pieces of legislation in the area of CSLs in the twentieth century.13 First, David Lloyd George’s Liberal government moved on the recommendations of the Lewis Re- port 1916 to raise the school leaving age from 13 to 14 as part of the post-WW1 reforms under the Education Act 1918 (or Fisher Act). The Act was ambitious in that it also aimed to institution- alize schooling until 18 and expand higher education, as well as abolish fees at state-run schools (although secondary education beyond the age of 14 did not become free until 1944) and establish a national schooling infrastructure. However, the change in the school leaving age was not implemented by Lloyd George until the Education Act 1921, coming into effect in 1922. Although the 1918 Act had intended for further increases in the leaving age, these did not transpire for ﬁnancial reasons despite repeated attempts in the 1920s and 1930s (Oreopoulos 2006). In practice, this Act had relatively little effect on school enrollment: most students remained in school until age 14. Staying in school beyond 14 typically required attending a grammar school (given secondary education was otherwise limited in supply and poorly subsidized), which entailed signiﬁcant school fees and passing an entrance exam. Consequently, the large majority of working class families did not send their children to secondary school. Second, as part of the Beveridge reforms, Churchill’s wartime coalition government passed the Education Act 1944 (or Butler Act), which increased the school leaving age from 14 to 15 in England and Wales;14 the Education (Scotland) Act 1945 cemented the same reform in Scotland. No such reform occurred in Northern Ireland until 1957, which is not included in the BES sur- 13This brief history borrows from Gillard(2011), Woodin, Gary and Cowan(2013, 2013) and the relevant legislative documents. 14The Education Act 1936 had determined that the age should be raised in 1939, but this did not occur because of the onset of WW2.

34 veys. The leaving age did not come into force until 1st April 1947, giving the education system time to expand its operations to accommodate the changes in the system (as well as many other new provisions under the 116-page monolith).15 However, the reform also established the new Tripartite system (coming into force in 1945), which meant that in most parts of the country formal secondary education began at 11 (rather than 14) and whether students attended a grammar, secondary technical or secondary modern school was determined by the “eleven plus” examination taken at age 11.16 As shown in the main paper, although fees had already been removed in 1944 and the new Tripartite system adopted in 1945, it was not until the schooling leaving age was raised that the dramatic increase in enrollment occurred (see also Oreopoulos 2006). The Education Act 1944 also provided for raising the leaving age to 16 once practical. Con- sequently the leaving age could be raised to 16 by an Order of Council.17 Conservative Prime Minister Harold Macmillan presided over plans to raise the school leaving age to 16 in the Edu- cation Act 1962, which ultimately fixed spring and summer leaving dates, although it was Con- servative Edward Heath who finalized the update to the current system under Statutory Instrument 444 (1972). The new rule, which had been overseen by Margaret Thatcher and heavily pushed by the Crowther Report 1959, was implemented for the academic year starting 1st September 1972 in England and Wales. Statutory Instrument 59 (1972) raised the leaving age more flexibly in Scot- land to allow local authorities, who were very concerned about teacher shortages (especially in Strathclyde/Glasgow), to allow part-time schooling and early leaving in the summer terms. Conse- quently, the 1972 reform was relatively weak for many Scottish students. The reform in Scotland was not fully implemented until the Education Act 1976. The Education (School-leaving Dates) Act 1976 introduced slightly more subtle leaving age rules—which are not utilized in this paper 15The lack of teachers was a serious concern, requiring an emergency training program in 1945 to address the lack of capacity. 16The Tripartite system was abolished in England in 1976 by the Labour government, and was replaced by the comprehensive system which did not bifurcate students at age 11. 17An Order of Council does not require approval like an Act. It may be lain before the House of Commons and is accepted unless a resolution is passed against it.

35 as they require monthly birth data (as in Clark and Royer 2013). In England, Wales and Scotland these reforms again raised education participation rates, although less dramatically than the 1944 Act (Milligan, Moretti and Oreopoulos 2004). Because the reform also introduced middle schools in some parts of the country, which meant that students entered secondary school one year later, the secondary population remained relatively steady. In 2008, Labour Prime Minister Gordon Brown passed the Education and Skills Act 2008. This requires that by 2013 young people must remain in at least part-time education or training until age 17; by 2015, this rises to 18. Although regional implementation will vary, the Act applies across the entire United Kingdom.

Variable deﬁnitions and summary statistics

All variables are from the British Election Survey (BES). Summary statistics for the main RD sample and the full BES sample (for which vote choice was available) are provided in Table4.

• Vote Conservative/Labour/Liberal. Indicator coded one for respondents who reported voting for the Conservative/Labour/Liberal Democrat party at the last general election. Note that Liberal party is used as a catch-all to include the Liberal Party, the Social Democrats in 1987 and the subsequent merged Liberal Democrats. Only respondents which refused to respond, did not answer or did not vote were excluded.

• Conservativel partisan. Indicator for identifying as a Conservative. The answer is to the following question: “Generally speaking, do you think of yourself as Conservative, Labour, Liberal, ...” (BES). Although a follow-up occurs if the respondent answers “none” or “don’t know”, this is treated as a zero in this analysis. Only respondents which refused to respond were excluded.

• Decided before campaign. Indicator coded one for respondents that reported that they had already devided who they would vote before the general election campaign.

36 Table 4: Summary statistics: RD and full BES samples

RD sample Full BES sample Obs. Mean Std. dev. Min. Max. Obs. Mean Std. dev. Min. Max. Dependent variable Vote Conservative 11,068 0.38 0.49 0 1 24,439 0.35 0.48 0 1 Vote Labour 11,068 0.36 0.48 0 1 24,439 0.36 0.48 0 1 Vote Liberal 11,068 0.19 0.39 0 1 24,439 0.19 0.39 0 1

Endogenous treatment variable Years of schooling 11,068 10.58 1.86 0 40 24,439 11.03 2.09 0 40

Excluded instrument Post 1947 reform 11,068 0.56 0.50 0 1 24,439 0.42 0.49 0 1

Pre-treatment covariates Survey year 11,068 1989.46 10.94 1974 2010 24,439 1990.85 11.13 1974 2010 Male 11,068 0.47 0.50 0 1 24,439 0.47 0.50 0 1 37 White 7,417 0.98 0.13 0 1 16,757 0.97 0.17 0 1 Black 7,417 0.01 0.07 0 1 16,757 0.01 0.09 0 1 Asian 7,417 0.01 0.08 0 1 16,757 0.01 0.12 0 1 Father manual/unskilled job 7,352 0.72 0.45 0 1 15,523 0.69 0.46 0 1 Age 11,068 55.54 13.10 27 91 24,439 49.56 17.59 18 98 Birth year 11,068 1933.92 8.31 1919 1947 24,439 1941.29 20.30 1877 1992

Mechanism variables Conservative partisan 10,424 0.37 0.48 0 1 22,844 0.34 0.47 0 1 Decided before campaign 11,068 0.74 0.44 0 1 24,439 0.68 0.47 0 1 Oppose tax and spend 7,370 3.35 2.39 0 10 17,188 3.48 2.34 0 10 Welfare gone too far 5,786 0.29 0.46 0 1 11,799 0.27 0.44 0 1 Oppose redistribution 7,256 1.69 1.23 0 4 15,463 1.66 1.22 0 4 Conservative economic policy scale 9,904 -0.02 1.01 -1.90 3.51 22,160 0.00 1.00 -1.90 3.51 Support crime reduction (over rights) 4,550 6.53 2.83 0 10 10,443 6.51 2.71 0 10 Support leaving Europe 9,181 0.36 0.48 0 1 20,854 0.32 0.47 0 1 Oppose abolishing private education 5,610 0.19 0.39 0 1 12,444 0.19 0.39 0 1 Political information index 6,097 0.12 0.93 -5.00 1.77 14,470 0.00 1.00 -5.00 1.77 Union member 10,138 0.26 0.44 0 1 22,011 0.25 0.43 0 1 • Years of schooling. Years of schooling is calculated as the age that the respondent left full time education minus ﬁve (the age at which students start formal schooling).

• Birth year. Birth-year is estimated by subtracting age at the date of the survey from the year in which the survey was conducted. Non-responses were deleted.

• Post 1947 reform. Indicator coded one for students aged 14 or below in 1947, and aged 15 or above in 1972.

• Male. Indicator coded one for respondents identifying as male. Non-responses were deleted.

• Age. Standardized age at the date of the survey.

• Race. Indicators coded one for respondents who respectively identify their ethnicity as white, black, or Asian (including South Asian ethnicities and Chinese).

• Father manual/unskilled job. Indicator coded one for respondent’s who answered that their father had a manual or unskilled job.

• Survey year. Year in which the survey was conducted.

• Oppose tax and spend. 11-point scale ranging from “Government should increase taxes a lot and spend much more on health and social services” (0) to “Government should cut taxes a lot and spend much less on health and social services” (10). Note that the scale in the raw data has been reversed. The 1974 and 1979 surveys did not ask this question, the 1983 survey used a 21-point scale that was re-scaled to 0-10, while the 1974 survey asked about just social services. Respondents that answered “don’t know” were excluded.

• Welfare beneﬁts gone too far. Indicator coded one for respondents that answer that welfare beneﬁts have “gone too far” or “gone much too far”. Not asked in 1997. Respondents that answered “don’t know” were coded as zero.

38 • Oppose redistribution. 4-point scale ranging from strongly agree (0) to strongly disagree (4) that the government should redistribute income and wealth.

• Conservative economic policy scale. Standardized summative rating scale combining the following three variables: oppose tax and spend, welfare beneﬁts gone too far, and oppose redistribution. Cronbach’s alpha of 0.42.

• Support crime reduction (over rights). 11-point scale ranging from government should pro- tect citizen rights (0) to government should emphasize reducing crime (10). Question was only asked in 1983 (rescaled from a 21-point scale), 1987, 2005, and 2010. Respondents that answered “don’t know” were excluded.

• Support leaving Europe. Indicator coded one for respondents that either support withdraw- ing from the EC, EEC or EU (1983, 1987, 1992, 1997 surveys) or disapprove or strongly disapprove of such membership (2001, 2005, 2010 surveys). Respondents that answered “don’t know” were coded as zero.

• Oppose abolishing private education. Indicator coded one for respondents that do not believe that private “deﬁnitely should”, “should” or “probably should” be abolished. Only available in the 1983-1997 surveys.

• Political knowledge index. Standardized scale (standardized within the RD sample) for the percentage of correct answers to a political information quiz from 1979, 1992 and 1997 surveys (standardized to also include the 2001-2010 elections). These surveys respectively asked 24, 11 and 7 correct-incorrect questions asking about party positions, politician recognition and knowledge of political institutions (e.g. voting age, parliamentary rules). The Cronbach’s alpha for each survey was very high. To ensure the scale is comparable across surveys, I created standardized mean scales for each survey and pooled across surveys.

• Union member. Indicator coded 1 for current members of trade unions or staff associations.

39 ifrnei est s006(0.071). 0.056 is density in is difference data the of density the that discontinuities. confirms reform 5 , the Fur- across Figure 2013). indistinguishable in Statistics shown National density, for ( 2008) (Office McCrary data around a rate heaping birth thermore, of annual lack by The confirmed is advance. reform in could 1947 decade parents the a since over reforms implausible CSL is predicted Britain precisely have in not at cohorts changes into simultaneously selection variable However, key another discontinuity. that the is designs RD for concern “sorting” key The checks assumption identifying RD

18 Density h ulhptei fadfeec ndniyars h icniut sfrfo eetd the rejected: from far is discontinuity the across density in difference a of hypothesis null The 0 .005 .01 .015 .02 1880 1900 otn test sorting (2008) McCrary 5: Figure 1920 Cohort: yearaged14 1940 40 18 1960 1980 2000 2020 Table 5: Continuity in other variables around the 1947 reform

Survey Male White Black Asian Father year manual/ unskilled job LLR LLR LLR LLR LLR LLR (1) (2) (3) (4) (5) (6) Post 1947 reform -0.101 0.002 0.002 0.004 -0.002 -0.044* (0.456) (0.021) (0.007) (0.004) (0.004) (0.023)

Observations 11,068 11,068 7,417 7,417 7,417 7,352

Notes: All speciﬁcations are local linear regressions using a triangular kernel and a bandwidth of 14.736. Robust standard errors in parentheses. * denotes p < 0.1, ** denotes p < 0.05, *** denotes p < 0.01.

Table5 confirms that pre-treatment variables do not differ significantly across the 1947 reform. In no case, is the coefficient on post 1947 reform close to being statistically significant. In each case I use the bandwidth used for the main results presented in the main paper.

Additional RD speciﬁcation tests

Columns (1)-(4) in Table6 shows that the results are robust to including quadratic or cubic polynomial trends in the running variable (birth year) either side of the discontinuity. Since I change the order of the polynomial, I also allow the Imbens and Kalyanaraman(2012) optimal bandwidth to adjust accordingly. Unsurprisingly, the estimates become less precise as higher-order polynomials are included. Columns (5) and (6) show the estimates when using the bias-corrected estimator proposed by Calonico, Cattaneo and Titiunik(2014). Finally, I examine placebo tests treating each of the ten years prior to the 1947 reform as a placebo reform. Figure6 presents the results of placebo reforms. Each bar is the estimate associated with treating the corresponding year as the 1947 reform. In no prior year, including those immediately before the reform, do we observe any evidence of an increase in Conservative

41 pia adit o ahpaeorfr.Br ersn out9%conﬁdence 95% robust represent Bars reform. placebo each for intervals. bandwidth optimal (2012) naraman Notes Effect of placebo reform on Vote Conservative -.1 -.05 0 .05 .1 1937 Ibn n Kalya- and Imbens the and kernel triangular a using regressions linear local are speciﬁcations All : 1938 iue6 feto lcb eom nCnevtv voting Conservative on reforms placebo of Effect 6: Figure 1939 1940 Placebo reformyear 1941 42 1942 1943 1944 1945 1946 Table 6: Robustness to higher-order polynomials in the running variable and bias correction

Vote Vote Vote Vote Vote Vote Con. Con. Con. Con Con. Con. LLR LLR IV LLR LLR IV LLR LLR IV (1) (2) (3) (4) (5) (6) Post 1947 reform 0.047* 0.055* 0.042** (0.025) (0.033) (0.019) Years of schooling 0.109* 0.124 0.098* (0.062) (0.080) (0.054)

Running variable polynomial order quadratic quadratic cubic cubic linear linear Optimal bandwidth 20.9 20.9 23.2 23.2 16.3 16.3 Observations 14,946 14,946 16,505 16,505 12,491 12,491 First stage F statistic 19.8 12.1 26.9

Notes: All speciﬁcations are local linear regressions using a triangular kernel. In columns (1)-(4) the bandwidth is the Imbens and Kalyanaraman(2012) optimal bandwidth, while the bandwidth in columns (5) and (6) is the Calonico, Cattaneo and Titiunik(2014) bias-corrected optimal bandwidth. Robust standard errors in parentheses. * denotes p < 0.1, ** denotes p < 0.05, *** denotes p < 0.01. voting. This provides strong support for the concern that the results are not being driven by pre- trend.

Exclusion restriction checks

Table7 presents the results of the exclusion restriction tests cited in the main paper. The estimates are from British Labor Force Surveys (LFSs). I use the summer quarter of LFS because this best matches when British elections are typically held and thus BES surveys occur. I then matched the pre-treatment sample moments to the BES sample to that the procedure produced relatively comparable estimates to the BES estimates in the main paper. In no case is there evidence of a signiﬁcant exclusion restriction violation.

43 Table 7: Exclusion restriction checks

Number of Age of Married dependent oldest at one children dependent time child (1) (2) (3) Post 1947 reform -0.020 -0.423 -0.009 (0.016) (0.462) (0.010)

Observations 9,712 881 5,058 Bandwidth 7.9 9.1 11.0

Notes: All estimates are from the July-September Labor Force Survey corresponding to the election year of each BES survey, and matched to the pre-treatment characteristics of the BES sample. All speciﬁcations are local linear regressions using a triangular kernel and the Imbens and Kalyanaraman(2012) optimal bandwidth for each sample. Robust standard errors in parentheses. * denotes p < 0.1, ** denotes p < 0.05, *** denotes p < 0.01.